反重力工作空间

1. Install GitNexus (requires Node.js)

npm install -g gitnexus

`ag-setup` — first-time configuration

Run this once per project, right after installing the plugin. Interactive picker for the LLM provider (OpenAI / DeepSeek / Groq / 阿里灵积 / NVIDIA NIM / Ollama local / any OpenAI-compatible endpoint), then writes .env to the project root with OPENAI_BASE_URL, OPENAI_API_KEY, OPENAI_MODEL, AG_ASK_TIMEOUT_SECONDS. Also ensures .env is in .gitignore. Skip it if you already have a working .env.

```

`ag-refresh` — build / refresh the knowledge base

Deploys the multi-agent cluster to read your code: each module gets its own Agent that produces a knowledge doc under .antigravity/agents/*.md, plus a map.md routing index. Run after install, after significant code changes, or when ag-ask returns stale answers. The first refresh auto-creates .antigravity/ — no separate init step needed. Pass quick for an incremental update, failed-only to rerun only previously failed modules.

```

Claude Code (auto-installs the Python engine CLI on first session via SessionStart hook)

/plugin marketplace add study8677/antigravity-workspace-template /plugin install antigravity@antigravity /antigravity:ag-setup # interactive: pick LLM provider, paste API key, writes .env /antigravity:ag-refresh # runs ag-refresh directly; first refresh auto-creates .antigravity/ /antigravity:ag-ask "How does this project work?" # runs ag-ask directly

Codex CLI (install the engine manually first; Codex hooks are not yet supported)

pipx install "git+https://github.com/study8677/antigravity-workspace-template.git#subdirectory=engine" codex plugin marketplace add study8677/antigravity-workspace-template /ag-setup # same flow, no antigravity: prefix in Codex /ag-refresh /ag-ask "How does this project work?"

Codex CLI auto-discovers slash commands from the plugin's `commands/` directory, so the same four commands work without the `antigravity:` namespace prefix (`/ag-setup`, `/ag-refresh`, `/ag-ask`, `/ag-init`). The raw CLI calls (`ag-refresh --workspace .`, `ag-ask "..." --workspace .`) also still work. If your Codex build supports MCP and you want tool-style integration, register `ag-mcp --workspace <project>` separately.

After install + setup you get `ag-ask <question>`, `ag-refresh`, and `ag-init <name>` slash commands in both hosts. MCP remains optional (`ask_project` + `refresh_project`) via `ag-mcp`; see [docs/examples/antigravity.mcp.json](docs/examples/antigravity.mcp.json). See [INSTALL.md](INSTALL.md) for details and troubleshooting.

**Option B — Manual install: engine + CLI via pip**

1. Install engine + CLI

pip install "git+https://github.com/study8677/antigravity-workspace-template.git#subdirectory=cli" pip install "git+https://github.com/study8677/antigravity-workspace-template.git#subdirectory=engine"

3. Build knowledge base (ModuleAgents self-learn each module)

ag-refresh --workspace .

Install both for full experience

pip install "git+https://...#subdirectory=cli" pip install "git+https://...#subdirectory=engine"

For local work on this repository itself:

---

Install engine

pip install "git+https://github.com/study8677/antigravity-workspace-template.git#subdirectory=engine"

Quick Start

Option A — Plugin install for Claude Code / Codex CLI ```bash

2. Configure .env with any OpenAI-compatible API key

cd my-project cat > .env <<EOF OPENAI_BASE_URL=https://your-endpoint/v1 OPENAI_API_KEY=your-key OPENAI_MODEL=your-model AG_ASK_TIMEOUT_SECONDS=120 EOF

5. (Optional) Register as MCP server for Claude Code

claude mcp add antigravity ag-mcp -- --workspace $(pwd)

**Option C — Context files only (any IDE, no LLM needed)**

CLI Commands

ag ask / ag refresh are available when both cli/ and engine/ are installed. ag-ask / ag-refresh are the engine-only entrypoints.

---

Two Packages, One Workflow

antigravity-workspace-template/
├── cli/                     # ag CLI — lightweight, pip-installable
│   └── templates/           # .cursorrules, CLAUDE.md, .antigravity/, ...
└── engine/                  # Multi-agent engine + Knowledge Hub
    └── antigravity_engine/
        ├── _cli_entry.py    # ag-ask / ag-refresh / ag-mcp + python -m dispatch
        ├── config.py        # Pydantic configuration
        ├── hub/             # ★ Core: multi-agent cluster
        │   ├── agents.py    #   Router + ModuleAgent + GitAgent
        │   ├── contracts.py #   Pydantic models: claims, evidence, refresh status
        │   ├── ask_pipeline.py    # agent.md + graph-enriched ask
        │   ├── refresh_pipeline.py # LLM-driven refresh → agents/*.md + map.md
        │   ├── ask_tools.py
        │   ├── scanner.py   #   multi-language project scanning
        │   ├── module_grouping.py # smart functional file grouping
        │   ├── structure.py
        │   ├── knowledge_graph.py
        │   ├── retrieval_graph.py
        │   └── mcp_server.py
        ├── mcp_client.py    # MCP consumer (connects external tools)
        ├── memory.py        # Persistent interaction memory
        ├── tools/           # MCP query tools + extensions
        ├── skills/          # Skill loader
        └── sandbox/         # Code execution (local / microsandbox)

CLI (pip install .../cli) — Zero LLM deps. Injects templates, logs reports & decisions offline.

Engine (pip install .../engine) — Repository knowledge runtime. Powers ag-ask, ag-refresh, ag-mcp. Uses the OpenAI-compatible endpoint written by ag-setup (OpenAI, DeepSeek, Groq, DashScope, NVIDIA NIM, Ollama, or custom).

New skill packaging updates: - engine/antigravity_engine/skills/graph-retrieval/ — graph-oriented retrieval tools for structure and call-path reasoning. - engine/antigravity_engine/skills/knowledge-layer/ — project knowledge-layer tools for semantic context consolidation.

```bash

Refresh knowledge base first (ModuleAgents self-learn each module)

ag-refresh --workspace /path/to/project

ModuleAgents self-learn your codebase

ag-refresh

Router intelligently routes to the right ModuleAgent

ag-ask "What testing patterns does this project use?"

Head-to-Head Eval: Antigravity vs Codex CLI vs Claude Code (2026-05-09)

Asymmetric benchmark on three real-world Python codebases — fastapi/fastapi, psf/requests, fastapi/sqlmodel — asking each tool the same 36 questions across three difficulty bands. All three tools used gpt-5.5 with high reasoning effort; Codex and Claude had full read access to the workspace. Codex was the grader (4-axis 0–3 rubric, scores verified against actual source).

Combined factual + audit (24 cells): Antigravity 284/288, Codex 283/288, Claude 276/288. Antigravity edges out both — at lower latency than Codex on every single question.

Latency (mean wall-clock per question, same proxy):

Antigravity is 2.1× faster than Codex on factual and on par with Codex on audit, while matching or beating it on correctness. Claude is fastest on audit but loses 7 percentage points of correctness.

What changed in this repo to get there. Two engine fixes landed during the benchmark, both committed in this branch:

1. _ask_with_agent_md now surfaces project-level docs (conventions.md, module_registry.md, map.md, structure.md) into its answer prompts. Removes the “module knowledge does not include project-wide conventions” refusal pattern. 2. The structured-facts answer agents now have search_code, read_file, list_directory, read_file_metadata, search_by_type bound at runtime, so the LLM can grep and read actual source instead of paraphrasing the KG.

Full report (data, methodology, per-cell tables, caveats): artifacts/benchmark-2026-05-09/REPORT.md.

---

Cross-IDE repository knowledge engine for grounded codebase Q&A.

ag-refresh builds the repository knowledge base. ag-ask routes questions to the right module context with source evidence. Plugins, CLI commands, and MCP are delivery channels around that core workflow.

Language: English | 中文 | Español

<br/>

<img src="https://img.shields.io/badge/Cursor-✓-000000?style=flat-square" alt="Cursor"/> <img src="https://img.shields.io/badge/Claude_Code-✓-D97757?style=flat-square" alt="Claude Code"/> <img src="https://img.shields.io/badge/Windsurf-✓-06B6D4?style=flat-square" alt="Windsurf"/> <img src="https://img.shields.io/badge/Gemini_CLI-✓-4285F4?style=flat-square" alt="Gemini CLI"/> <img src="https://img.shields.io/badge/VS_Code_+_Copilot-✓-007ACC?style=flat-square" alt="VS Code"/> <img src="https://img.shields.io/badge/Codex-✓-412991?style=flat-square" alt="Codex"/> <img src="https://img.shields.io/badge/Cline-✓-FF6B6B?style=flat-square" alt="Cline"/> <img src="https://img.shields.io/badge/Aider-✓-8B5CF6?style=flat-square" alt="Aider"/>

</div>

<br/>

<br/>

`ag-ask` — routed Q&A on the codebase

The main reason this plugin exists. Routes your question to the right ModuleAgent (and GitAgent / GitNexus when applicable), then returns an answer grounded in actual source with file paths and line numbers. Use it before manually grepping or reading files — it's faster and more accurate. Good question shapes: "where is X defined/handled?", "why was Y done this way?", "how does the auth flow work?", "what depends on module Z?".

```

3. `ag-ask` — Router-based Q&A

ag-ask "How does auth work in this project?"

The ask pipeline uses a dual-path architecture: - Semantic path: Router reads map.md → selects modules → reads agents/*.md → LLM answers with code references. Multiple agent docs are read in parallel, then a Synthesizer combines answers. - Graph path (automatic): Router LLM decides if the question needs structural analysis → queries GitNexus for call chains, dependencies, or impact → injects graph data into the answer context. Silently skipped if GitNexus is not installed.

Falls back to the legacy Router → ModuleAgent/GitAgent swarm when agent docs are not yet generated.

---